Els Vrindts

Weed Detection Using Canopy Reflection

For site-specific application of herbicides, automatic detection and evaluation of weeds is desirable. Since reflectance of crop, weeds and soil differs in the visual and near infrared wavelengths, there is potential for using reflection measurements at different wavelengths to distinguish between them. Reflectance spectra of crop and weed canopies were used to evaluate the possibilities of weed detection with reflection measurements in laboratory circumstances. Sugarbeet and maize and 7 weed species were included in the measurements. Classification into crop and weeds was possible in laboratory tests, using a limited number of wavelength band ratios. Crop and weed spectra could be separated with more than 97% correct classification. Field measurements of crop and weed reflection were conducted for testing spectral weed detection. Canopy reflection was measured with a line spectrograph in the wavelength range from 480 to 820 nm (visual to near infrared) with ambient light. The discriminant model uses a limited number of narrow wavelength bands. Over 90% of crop and weed spectra can be identified correctly, when the discriminant model is specific to the prevailing light conditions.

A neural network based plant classifier

The Self-Organizing Map (SOM) neural network is used in a supervised way for a classification task. The neurons of the SOM are extended with local linear mappings. Error information obtained during training is used in a novel learning algorithm to train the classifier. The proposed method achieves fast convergence and good generalization. The classification method is then applied in a precision farming application, the classification of crops and weeds using spectral properties. The proposed method compares favorably with an optimal Bayesian classifier that is presented in the form of a probabilistic neural network. The classification performance of the proposed method is proven superior compared with other statistical and neural classifiers.

Analysis of Soil and Crop Properties for Precision Agriculture for Winter Wheat

In a precision farming research project financed by the Belgian Ministry of Small Trade and Agriculture, the methods of precision agriculture are tested on grain fields with a view of implementation of precision agriculture methods in Belgian field agriculture. The project encompasses methods for automatic information gathering on soil and crop and analysis of this data for management of within-field variability. Automatic information capturing is combined with traditional data sources of soil sample analysis and crop observations. The measurements and part of the results on one particular field in Sauveniere are presented here. Five nitrogen management strategies were compared, but the resulting differences in nitrogen dose were small and did not lead to significantly different yield results. The yield results were correlated to topography-related variations in soil texture and chemical components and to crop reflectance measurements in May.

Fine-scaled optical detection of nitrogen stress in grain crops

What is lacking in precision farming at present are more comprehensive and fast non-destructive methods for obtaining the data needed to prescribe varia*ble treatments. In precision farming there is a demand for sensors that can easily monitor crop nitrogen requirements throughout the growing season on a high resolution. Currently used optical measurement platforms like satellites, airplanes and hand-held sensors, do not meet the needs of precision agriculture for good nitrogen management possibilities. An automated sensor system mounted on a tractor was developed and used to detect crop nitrogen status optically. A line spectrograph was used to detect amount of nitrogen (kgN/ha) and chlorophyll (kg/ha) in a wheat crop (Triticum aestivum L.). By calculating the red edge inflection point of the plant spectra in the images, wheat crop nitrogen stress within small areas in the field could be detected. Spectrograph red edge was highly correlated with applied nitrogen to the wheat crop (0.90), with crop nitrogen uptake (0.89) and with chlorophyll amount in the crop (0.86). The average errors when estimating those variables with the red edge inflection point were -0.4% (24.15kgN/ha), -1% (17.25kgN/ha) and -10% (14.74kg/ha) respectively. This means that spectrograph red edge measurements of the wheat crop during the growing season can be a predictor of topdress nitrogen needs.

Optical weed detection and evaluation using reflection measurements

For the site-specific application of herbicides, the automatic detection and evaluation of weeds is necessary. Since reflectance of crop, weeds and soil differs in visual and near IR wavelengths, there is a potential for using reflection measurements at different wavelengths to distinguish between them. Diffuse reflectance spectra of crop and weed leaves were used to evaluate the possibilities of weed detection with reflection measurements. Fourteen different weed species and four crops were included in the dataset. Classification of the spectra in crop, weeds and soil is possible, based on 3 to 7 narrow wavelength bands. The spectral analysis was repeated for reflectance measurements of canopies. Sugarbeet and Maize and 7 weed species were included in the measurements. The classification into crop and weeds was still possible, suing a limited number of wavelength band ratios. This suggest that reflection measurements at a limited number of wavelength bands could be used to detect and treat weeds in a field. This is a great environmental benefit, as agrochemicals will only be used where they are needed. The possibilities of using optical reflectance for weed detection and treatment in the field are discussed.

Analysis of Spatial Soil, Crop and Yield Data in a Winter Wheat Field

In 2001 and 2002, soil and crop parameters were measured on a winter wheat field in Sauveniere, as part of a precision farming research project. One of the objectives was to study the processing of precision farming data for correct use in precision management. Different methods to study the relation between soil and crop were tested: correlation analysis, principal component analysis of soil parameters and clustering of soil and yield parameters. Crop and soil data were interpolated to a 6m grid and a 10 m grid to check the effect of grid size on the correlations between field data. Correlations were very similar for the 2 datasets, with slightly higher values for the 10 grid data (difference of 0 to 0.02 in correlation values). Grain and straw yield in 2001 were correlated to soluble phosphate, texture parameters, soil electrical conductivity, and potassium (coefficient of determination R² values up to 0.30 for grain yield). Crop optical measurements in May 2001 had lower correlation to soil parameters than yield (coefficient of determination R²= 0 to 0.18). Correlations were higher in March 2002, with coefficient of determination values up to 0.66 for optical mesurements. Correlation of grain yield to soil was very low in 2002, in part because of the incidence of eye spot disease. Principal component analysis of soil parameters resulted in three principal components describing the overall soil texture variation over the field, soil organic matter and soil nitrogen in early spring and acidity and phosphate variability. Correlations between yield and crop measurements and soil principal components were as expected from the correlation analysis. Clustering soil parameters resulted in soil zones that did not coincide with crop variability. Clustering yield and soil electrical conductivity did lead to zones that could be used to set up management zones. The average soil properties of these zones could be used to find parameters linked to yield variability and as a start to determine causes of variability in crop growth and yield, using a broader knowledgde on the soil-plant interaction. The grid size influenced the results of the analysis and should be further investigated, to determine the best methods for processing precision farming data. The soil data collected from the top 30 cm layer and the soil electrical conductivity only explain a limited part of variability in crop growth and yield. This means that either the top 30 cm was not representative of growth conditions or that other, non-measured soil parameters were important for crop growth. Root depth, water availability and soil compaction were probably important for the crop growth in 2001.

Use of very high resolution satellite images for precision farming: recommendations on nitrogen fertilization

combined with land parcel data. As a part of this prototype system, an Algorithm provides tools for the partial automation ofthe decision process to formulate N-fertilisation recommendations, in the form of a Nitrogen Application Map. The mostimportant part of the algorithm is the work unit based on very high resolution remotely sensed data. To develop this workunit, field measurements were taken parallel to satellite images during two growing seasons (1999-2000) on winter wheatparcels with different plant densities and nitrogen treatments. The parameters measured included reflectance and Leaf AreaIndex (LAI) during the growing season and yield, protein content and number of spikes, at harvest.Several vegetation indices (Vis), calculated from the satellite and ground data, were studied with respect to their sensitivityto the different nitrogen doses and to low noise production. The correlation between Vis and important crop parameters fornitrogen fertilisation management (i.e. LAI) was tested as well.Keywords: Remote Sensing, Canopy Management, Precision Agriculture, Very High Resolution Satellites

Precision farming through variable fertilizer application by automated detailed tracking of in-season crop properties

What is lacking in precision farming at present, are more comprehensive and non-destructive methods for obtaining the data needed to prescribe variable treatments. A farmer needs to be informed in order to be efficient, and that includes having the knowledge and information products to forge a viable strategy for farming operations. Current remote sensing (satellite images) sources are too coarse in multispectral spatial resolution and too infrequent in time to allow detailed tracking of phenological stages during the growing season. In this research very detailed and automated on-the-go optical monitoring of the crop is used for detecting and managing zones with different crop yield potential on a seasonal scale. In particular, reflectance properties are used to identify and evaluate optical indicators of the nutritional status of the crop. These indicators should allow site-specific in-seasonal correction of N-application to come to optimal crop yield all over the field. Based on these indicators, site-specific fertilization is done with a variable fertilizer equipped with DGPS. At the end of the season, the crop was harvested with a combine harvester, equipped with precision farming sensors to map final crop yield. In this way final results could be evaluated and analyzed.